what if?
zkLTS is the key to providing trustless LLM inference since it can't be done on chain
Smart contracts!
To make a profit, we invest collected premiums. Profits are given back to restakers and operators!
try at https://eigensurance.replit.app/

Video only shows one part, please see What I Made section + click thru slideshow

diagram

Inspiration 💡🏠

Coming from Southern California, I'm very aware of the immense risks posed by wildfires, like the ones in LA last month. In fact, there was a brush fire just a few miles from my house. Unfortunately, increasing wildfire risks means that most home insurers are leaving the state, leaving millions without insurance. And, the remaining insurers have increased prices significantly.

I've never done any crypto/web3 stuff before but wanted to explore for the hackathon. After reading about decentralized projects like Filecoin and Helium, I wondered: could we do the same for insurance? Instead of one profit-oriented corporation, services could be rendered transparently, quickly, and fairly on the blockchain, assisted by AI agents.

I realized that insurance companies basically have three jobs:

Risk assessment: determining how much insurance costs based on risk models
Claims processing: reviewing claims and disbursing/denying funds
Treasury investing: insurance companies make profit by investing collected money

Having humans do this is very expensive. ~27% of annual premiums collected ($560B) is spent on operating expenses like salaries, sales commissions, management fees, etc.

I realized that all three of these components could be replaced by AI and web3 technology, driving down costs exponentially—meaning that insurance can become cheaper, faster, and more fair. All this is enabled by the** Eigenlayer AVS system**, which enables us to get zkTLS verified off-chain LLM responses through Opacity and Cloudflare. Eigenlayer provides a way for independent operators to provide services like LLM inference, and they are incentivized to be truthful because they get slashed (lose their stake) if they are caught lying. Also, by being on the blockchain, we can ensure transparent pricing and payout conditions via smart contracts.

What I made 🔗

While I wasn't able to finish the entire thing due to the complexity of the project and working solo, I was able to develop various working + open sourced key components including an AI agent and various smart contracts.

a landing page + detailed technical whitepaper
on a local testnet based on Eigenlayer Hello World AVS, a smart contract setup where users can purchase insurance policies, process claims, and simulate disaster events for testing. Operators would provide verified LLM inference, which powers the agents. Github repo
local demo that uses Opacity and Perplexity API to review claims and verify the disaster events via Perplexity API, and returns a zkTLS-verified verdict. I hard-coded in satellite, weather, and other data for the time being but the Opacity GPT4o zkTLS stuff is real and so is the Perplexity API! Github repo
more complex insurance policy smart contract with features like voting, various levels of insurance, expiry, etc.
smart contract for wildfire prediction market

**To invest collected premiums to make profit to reward restakers/operators and pay policyholders:

Financial AI agent (uses five tools) that invests premiums using real data + calculator + submits simulated trade orders. Live demo Advanced version with simulated trades, and Github

How I built it 🛠️

First of all, it was only possible thanks to a lot of explanations and help from the Eigenlayer and Opacity teams since this is my first time building a web3 project.

To build things, I relied on the sample repos, Eigenlayer Hello World AVS and the Eigenlayer AI quickstart. The main languages were Solidity, TypeScript, and some HTML/CSS for the frontends.

I also used Remix which is kind of like Replit for web3, which helped me test my smart contracts. I also used Google Colab to code and test the stock trading agent and then Replit to deploy the live demo.

Challenges I ran into

Learning the intricacies of smart contracts, on‑chain/off‑chain interactions, and integrating AVS Getting set up with a MetaMask wallet Remembering how much gwei, wei, and eth are